MUMBAI, India, Oct. 26 -- Intellectual Property India has published a patent application (202641123779 A) filed by Vellore Institute Of Technology on October 15, 2026, for Method For Multi-Granular Deletion And Concurrent Data Ingestion In Sharded Machine Unlearning.
Inventors include Ramalingam Murugan; and Rajeswari C.
The application for the patent was published on October 23, 2026, under issue no. 43/2026.
Abstract: ABSTRACT METHOD FOR MULTI-GRANULAR DELETION AND CONCURRENT DATA INGESTION IN SHARDED MACHINE UNLEARNING A computer-implemented machine unlearning method partitions records of a plurality of data subjects in a training dataset (102) into disjoint shards (112), each data subject's record being assigned to exactly one shard (112). Each shard (112) is partitioned into sequential slices, and a shard model (114) is trained per shard in isolation, slice by slice, storing a checkpoint in a checkpoint store (116) after each slice. An aggregation module (118) aggregates shard model (114) outputs into a final prediction (120). A hierarchical indexing module (108) indexes each sub-record by a composite key of a data-subject identifier, a sub-record identifier, and a slice identifier. Upon a deletion request (126) specifying sub-record-level scope, the identified sub-record is removed while other sub-records and other data subjects' data are retained, and the affected shard model (114) is retrained from the checkpoint preceding the keyed slice using the retained sub-records. (Fig. 1)
Disclaimer: Curated by HT Syndication.
